Hi,
I need some advice on the following. I tried to implement the B-spline
kernel as defined in one of the first paper of Vapnik on this subject:
'Support Vector Method for function approximation, regression estimation,
and signal processing'. When implemented as given, I notice a strange
behaviour (numerical instability, not symmetric) for negative values.
Especially for higher order (e.g.. 41) B-spline, this instability increases.
Has someone else also noticed this problem? Can you perhaps give some
advice or references about how to implement these B-spline in a numerical
stable way.
Thanks in advance.
Bart.
____________________________________________
Bart Hamers
Research Assistant
University of Leuven
Department of Electrical Engineering ESAT
SCD-SISTA
Kasteelpark Arenberg 10
3000 Leuven
T: 0032/16 321145
F: 0032/16 321970
E: [log in to unmask]
W: http://www.esat.kuleuven.ac.be/~bhamers/
____________________________________________
|